This is a really good point.

Related, Matt Levine has recently been writing about how the AI labs are paying huge amounts of money to tech/math/AI people but these people don't actually know how to spend it. https://www.bloomberg.com/opinion/newsletters/2026-09-10/dal... As opposed to the typical finance worker who knows very well how to spend.

Yes, you need to manually verify the statement of the theorem of interest of formalized correctly. But you don't need to anything more than this: you can rely on the proof being correct. And the proof is overwhelmingly the most amount of code.

>you don't need to anything more than this

You also have to check for things like sorry or defining axioms.

Yeah I think the verb "removed" is not the right verb here, because the paper in question is OpenAI's hypothetical paper which Alpöge is not on in the first place.

NYC zoning is less limiting than LA, but still extremely limiting.

I agree; that’s why I think we need a moratorium on human drivers while we figure this all out.

We do tend to impose them on individual drivers who make egregious and consequential failures, that’s the point. If every car running the same software is functionally the same ‘driver’ this class of failure absolutely should trigger a moratorium.

I think this is a fantastic distinction to make, and one I haven’t seen before.

What is this Tesla (or its non-driver) going to get?

more retail investors buying the stock

This also solves the rising oil and gas prices issue as well!

Very nice. As a follow up would be interesting to also reverse engineer the structure of the vtable itself. I guess it’s a list of pointers to the method implementations?

Not just pointers, it also has info such as alignment. (Shameless plug) I wrote a relevant article recently: https://ashdnazg.github.io/articles/26/Downcasting-Arcs-in-R...

Awesome, thank you for sharing!

My first question was, if the vtable only has a single function in it (as in these examples), can the compiler optimize it to a function pointer to save an indirection?

The vtable contains at least the following informations:

- the size of the pointed type - the alignment of the pointed type - a function pointer for its drop glue

That's already quite big to inline to save on indirection, and this is before adding more methods.

I think the issue is that you knew there was a bug to be found in the write and checkpoint interactions, which then determined which workload you built.

My understanding is that this workload is very uncommon: the Tailscale blog says they used a custom unusual configuration to have many checkpoints like this. So without knowing the bug, it seems unlikely one would build this workload and then find the bug. If that makes sense.

Edit: just want to say that you being able to repro it is awesome, but that the overall claim seems a little overstated to me.

I can see how the post comes across that way. I may need to edit it to somehow be clearer about how I approached the workload development. I'll refer you to my other replies to comments for more information on my approach:

https://news.ycombinator.com/item?id=49278424 https://news.ycombinator.com/item?id=49278521

Also appreciate the nice words at the end :) I'm feeling a bit ganged up on.

I do feel like you're being ganged up on as well. IMO this is a very cool thing that Antithesis is able to find and reproduce this bug in this specific subsystem with a relatively generic workload.

I guess what would have been an even more cool thing is "we ran some more general testing with Antithesis, and it found five other bugs". Have you thought along those lines or explored something like that? There have to be other, similar bugs lurking in SQLite :)

I’m sorry you’re feeling ganged up on, I thought it was a nice article and cool you were able to repro the bug with your tools.

Like some others have mentioned, one of my earliest thoughts was “how much of a hint was the LLM given about the bug?” I think if the prompt used was stated clearly/verbatim near the beginning of the article, that would probably dispel a good amount of the criticism.

I disagree in the sense that you can rewrite the code to use the trick and also not allocate in advance. Nothing about the trick requires you to allocate up front: before writing to out[n] you can extend the vector if it’s out of bounds. Or, after incrementing n, do out.push(0).

You should try writing it out. Doing it without introducing another unpredictable branch is harder than it looks.

I discussed this with a coworker earlier this week and the best they were able to come up with was

    for &x in input {
        out.push(x);
        n += (x > threshold) as usize;
        out.truncate(n);
    }

which works but is ugly af imo.

Yep realized this after that my second solution (push a 0 if n is incremented) has the same branch prediction problem.

I think yours works. Alternatively in the loop:

    if out.len() < n {
        out.push(0);
    }
    out[n] = x;
    n += (x > threshold) as usize;

In this case the if will be predicted well because it only triggers log(N) times, given how the std lib extends vectors.

Consider turning off the internet before you go to bed and not turning it on for at least 1 hour after you wake up. I think I got this idea from Craig Mod (who I believe does no internet until after lunch). I've been doing this for a few months. The initial hour is really good (very easy to e.g. read or work calmly on some software projects without distraction) but I also find it puts me in the right mindset for the rest of the day.

Kind of hard if you work 9-5 need to know what's gonna happen before you pull up to the office. Do I have a meeting at 9 AM kind of thing.

Yes the night before you look at your calendar and figure out what time you need to be online at.

In general if you want a no-internet 1 hour in the morning, organize yourself to wake up 1 hour earlier (maybe going to sleep 1 hour earlier). You sleeping until 8am, and you sleeping until 7am and having no internet for 1 hour, is equivalent to the outside world.

This information about the NYT is out of date; it's now easy to unsubscribe from anywhere.

And that is courtesy of recent New York State law. [1]

[1] https://kr.law/news/article-detail/new-yorks-new-automatic-r...

Nice!